Step 2: Write and examine the code - Amazon Kinesis Video Streams
Services or capabilities described in Amazon Web Services documentation might vary by Region. To see the differences applicable to the China Regions, see Getting Started with Amazon Web Services in China (PDF).

Step 2: Write and examine the code

In this section of the C++ Producer Library procedure, you examine the code in the C++ test harness (tst/ProducerTestFixture.h and other files). You downloaded this code in the previous section.

The Platform Independent C++ example shows the following coding pattern:

  • Create an instance of KinesisVideoProducer to access Kinesis Video Streams.

  • Create an instance of KinesisVideoStream. This creates a Kinesis video stream in your Amazon Web Services account if a stream of the same name doesn't already exist.

  • Call putFrame on the KinesisVideoStream for every frame of data, as it becomes available, to send it to the stream.

The following sections provide more information about this coding pattern.

Creating an instance of KinesisVideoProducer

You create the KinesisVideoProducer object by calling the KinesisVideoProducer::createSync method. The following example creates the KinesisVideoProducer in the ProducerTestFixture.h file:

kinesis_video_producer_ = KinesisVideoProducer::createSync(move(device_provider_), move(client_callback_provider_), move(stream_callback_provider_), move(credential_provider_), defaultRegion_);

The createSync method takes the following parameters:

  • A DeviceInfoProvider object, which returns a DeviceInfo object containing information about the device or storage configuration.


    You configure your content store size using the deviceInfo.storageInfo.storageSize parameter. Your content streams share the content store. To determine your storage size requirement, multiply the average frame size by the number of frames stored for the max duration for all the streams. Then multiply by 1.2 to account for defragmentation. For example, suppose that your application has the following configuration:

    • Three streams

    • 3 minutes of maximum duration

    • Each stream is 30 frames per second (FPS)

    • Each frame is 10,000 KB in size

    The content store requirement for this application is 3 (streams) * 3 (minutes) * 60 (seconds in a minute) * 10000 (kb) * 1.2 (defragmentation allowance) = 194.4 Mb ~ 200 Mb.

  • A ClientCallbackProvider object, which returns function pointers that report client-specific events.

  • A StreamCallbackProvider object, which returns function pointers that are called back when stream-specific events occur.

  • A CredentialProvider object, which provides access to Amazon credential environment variables.

  • The Amazon Web Services Region ("us-west-2"). The service endpoint is determined from the Region.

Creating an instance of KinesisVideoStream

You create the KinesisVideoStream object by calling the KinesisVideoProducer::CreateStream method with a StreamDefinition parameter. The example creates the KinesisVideoStream in the ProducerTestFixture.h file with the track type as video, and with track id as 1:

auto stream_definition = make_unique<StreamDefinition>(stream_name, hours(2), tags, "", STREAMING_TYPE_REALTIME, "video/h264", milliseconds::zero(), seconds(2), milliseconds(1), true, true, true); return kinesis_video_producer_->createStream(move(stream_definition));

The StreamDefinition object has the following fields:

  • Stream name.

  • Data retention period.

  • Tags for the stream. These tags can be used by consumer applications to find the correct stream, or to get more information about the stream. The tags can also be viewed in the Amazon Web Services Management Console.

  • Amazon KMS encryption key for the stream. For more information, see Using Server-Side Encryption with Kinesis Video Streams.

  • Streaming type. Currently, the only valid value is STREAMING_TYPE_REALTIME.

  • Media content type.

  • Media latency. This value isn't currently used, and should be set to 0.

  • Playback duration of each fragment.

  • Media timecode scale.

  • Whether the media uses key frame fragmentation.

  • Whether the media uses timecodes.

  • Whether the media uses absolute fragment times.

Adding an audio track to the Kinesis video stream

You can add audio track details to a video track stream definition by using the addTrack method of the StreamDefinition:


The addTrack method requires the following parameters:

  • Track id (as one for audio). This should be unique and non-zero value.

  • User-defined track name (for example, "audio" for the audio track).

  • Codec id for this track (for example, for audio track "A_AAC").

  • Track type (for example, use the enum value of MKV_TRACK_INFO_TYPE_AUDIO for audio).

If you have codec private data for the audio track, then you can pass it when calling the addTrack function. You can also send the codec private data after creating the KinesisVideoStream object while calling the start method in KinesisVideoStream.

Putting a frame into the Kinesis video stream

You put media into the Kinesis video stream using KinesisVideoStream::putFrame, passing in a Frame object that contains the header and media data. The example calls putFrame in the ProducerApiTest.cpp file:

frame.duration = FRAME_DURATION_IN_MICROS * HUNDREDS_OF_NANOS_IN_A_MICROSECOND; frame.size = SIZEOF(frameBuffer_); frame.frameData = frameBuffer_; MEMSET(frame.frameData, 0x55, frame.size); while (!stop_producer_) { // Produce frames timestamp = std::chrono::duration_cast<std::chrono::nanoseconds>( std::chrono::system_clock::now().time_since_epoch()).count() / DEFAULT_TIME_UNIT_IN_NANOS; frame.index = index++; frame.decodingTs = timestamp; frame.presentationTs = timestamp; // Key frame every 50th frame.flags = (frame.index % 50 == 0) ? FRAME_FLAG_KEY_FRAME : FRAME_FLAG_NONE; ... EXPECT_TRUE(kinesis_video_stream->putFrame(frame));

The preceding C++ Producer example sends a buffer of test data. In a real-world application, you should obtain the frame buffer and size from the frame data from a media source (such as a camera).

The Frame object has the following fields:

  • Frame index. This should be a monotonically incrementing value.

  • Flags associated with the frame. For example, if the encoder were configured to produce a key frame, this frame would be assigned the FRAME_FLAG_KEY_FRAME flag.

  • Decoding timestamp.

  • Presentation timestamp.

  • Duration of the frame (to 100 ns units).

  • Size of the frame in bytes.

  • Frame data.

For more information about the format of the frame, see Kinesis Video Streams Data Model.

Putting a KinesisVideoFrame into a specific track of KinesisVideoStream

You can use the PutFrameHelper class to put frame data into a specific track. First, call the getFrameDataBuffer to get a pointer to one of the pre-allocated buffers to fill in the KinesisVideoFrame data. Then, you can call the putFrameMultiTrack to send the KinesisVideoFrame along with the Boolean value to indicate the type of frame data. Use true if it's a video data or false if the frame contains audio data. The putFrameMultiTrack method uses a queueing mechanism to ensure that the MKV Fragments maintain monotonically increasing frame timestamps and any two fragments don't overlap. For example, the MKV timestamp of the first frame of a fragment should always be greater than the MKV timestamp of the last frame of the previous fragment.

The PutFrameHelper has the following fields:

  • Maximum number of audio frames in the queue.

  • Maximum number of video frames in the queue.

  • Size to allocate for a single audio frame.

  • Size to allocate for a single video frame.

Metrics and metric logging

The C++ Producer SDK includes functionality for metrics and metric logging.

You can use the getKinesisVideoMetrics and getKinesisVideoStreamMetrics API operations to retrieve information about Kinesis Video Streams and your active streams.

The following code is from the kinesis-video-pic/src/client/include/com/amazonaws/kinesis/video/client/Include.h file.

/** * Gets information about the storage availability. * * @param 1 CLIENT_HANDLE - the client object handle. * @param 2 PKinesisVideoMetrics - OUT - Kinesis Video metrics to be filled. * * @return Status of the function call. */ PUBLIC_API STATUS getKinesisVideoMetrics(CLIENT_HANDLE, PKinesisVideoMetrics); /** * Gets information about the stream content view. * * @param 1 STREAM_HANDLE - the stream object handle. * @param 2 PStreamMetrics - Stream metrics to fill. * * @return Status of the function call. */ PUBLIC_API STATUS getKinesisVideoStreamMetrics(STREAM_HANDLE, PStreamMetrics);

The PClientMetrics object filled by getKinesisVideoMetrics contains the following information:

  • contentStoreSize: The overall size in bytes of the content store (the memory used to store streaming data).

  • contentStoreAvailableSize: The available memory in the content store, in bytes.

  • contentStoreAllocatedSize: The allocated memory in the content store.

  • totalContentViewsSize: The total memory used for the content view. The content view is a series of indices of information in the content store.

  • totalFrameRate: The aggregate number of frames per second across all active streams.

  • totalTransferRate: The total bits per second (bps) being sent in all streams.

The PStreamMetrics object filled by getKinesisVideoStreamMetrics contains the following information:

  • currentViewDuration: The difference in 100 ns units between the head of the content view (when frames are encoded) and the current position (when frame data is sent to Kinesis Video Streams).

  • overallViewDuration: The difference in 100 ns units between the head of the content view (when frames are encoded) to the tail (when frames are flushed from memory, either because the total allocated space for the content view is exceeded, or because a PersistedAck message is received from Kinesis Video Streams, and frames known to be persisted are flushed).

  • currentViewSize: The size in bytes of the content view from the head (when frames are encoded) to the current position (when frames are sent to Kinesis Video Streams).

  • overallViewSize: The total size in bytes of the content view.

  • currentFrameRate: The last measured rate of the stream, in frames per second.

  • currentTransferRate: The last measured rate of the stream, in bytes per second.


If you want to send the remaining bytes in a buffer and wait for the ACK, you can use stopSync:


Or you can call stop to end the streaming:


After stopping the stream, you can free the stream through invoking the following API:


Next step

Step 3: Run and verify the code